Johannes Weiss

Thanks @Lukasa, that all makes sense. I personally wouldn't dupe it because at least to me it was news that GAI doesn't RR and this might be a problem that...

@PeterAdams-A We're using `pthread_mutex_t` at the moment (on the heap) which is a fair lock on Darwin (unfair on Linux). We want unfair. @artemredkin and I have actually done some...

ConditionLock should stay on pthread, just Lock should use unfair_lock

It does today because it made the implementation easier and ConditionLock can just reach into the implementation details of Lock. For layering reasons this isn’t ideal anyway but so far...

I think on Darwin `pthread_mutex_t` is now itself backed by a `os_unfair_lock` but this needs to be checked. Maybe this can be closed.

whoops :)

> Hmm, I agree that the pattern is ugly, but do we think it rises to "really bad" in a way that justifies having a singleton event loop group? I...

> I mean, the simplest idea is the one that we have been falling back to all along: creating the ELG is the app's business, not the business of any...

@Lukasa or we could have a new repo, let's call it `swift-nio-family` which has a `NIOGlobalExecutor` or so module or whatever which depends on everything and provides an ELG that...

1. Correct 2. This is already the case today. Libraries like AHC/gRPC choose NIO on Sockets or NIOTS but they don't just choose others. I'm very open to better, more...